Integrate OpenAI assistant

General Discussion
1

We absolutely need an easy integration of OpenAI assistant into Adalo, I was able to integrate it, but its a bit of a hit and miss, it works flawless, and sometimes it doesn’t, would love if we can get just a no-code integration lol like you did for GPT.

Thanks

2

Hey, when you say it doesn’t work what happens? How have you integrated? Using the OpenAi API? or Through make? Are you URLEncoding your inputs before sending as your prompts? (API calls do not like special Characters and white space - this is the most common reason for the call to fail).

The other thing is that since with the Open AI assistants the vector databasing is being taken care of by OpenAI/Azure OpenAI there will always be two Parts 1. User Sends Message to the Thread, 2. Run thread. As the “runs thread part” doesn’t send a response back the same way the completion API does, you then have to incrementally check the “status” of the thread run, and a lot of the assistants API calls can take 40+ seconds, which then times out Adalo’s wait time.

I have integrated and works fairly consistently. The only times it doesnt work is if the API is down itself… Which since moving to Azure OpenAI API is very rare.

A really simple method is use Make.com (which elmiates the “checking status part of the Run”.

Setup:

In Adalo
Collection: Conversations (with Thread ID Field)
Collection: Messages (with relationship to Conversation) 1 > Many

  1. Send a Webhook to Make.com with “Adalo Conversation ID”, “Thread ID”, “Assistant ID” and “Prompt” (assuming you have an assistant set up in OpenAi) (Make sure you URLEncode the Prompt)
  2. Add the Open Ai Module in Make (dynamically link the Thread ID, Assistant ID & Prompt if you are mapping these.
  3. Add an Adalo Module “Create Record” with the Ai response and the Conversation ID in the 1 > Many relationship field
  4. Add another “Webhook Response” by selecting Raw on the Webhook, and pass the below Json in the body of the webhook response:

{
“Response”: “ReplaceTextWithMakeOpenAiResponse”
}

The big thing about this method in Make is that it does it all in one button click and elimiates the “Check Status” of the run. Most of the time, the response comes back within 20-30 seconds and the Adalo button spinner keeps spinning until you get a response. Worst case, if the “Spinning” Adalo button stops due to Adalo timeout, then when you have Autorefresh set on your conversation in Adalo the message will appear (you can also set up status loaders/spinners etc for better UX in Adalo while waiting).

1 Like
3

Thank you for this, This is pretty cool, I ll try it , i m a big fan of doing all inhouse lol, using Adalo only, but I guess its not possible now as I have a send prompt, and retrieve response now, and as you said, sometimes the spining times out leading the user to confusion. Is the setup a limitation of Adalo? as I see other users setup assistant fine like a chatbot in other platforms or pure code…

Question: I m not sure I understand the Autorefresh part, where do you this? Adalo or Make?

image
image1271×194 2.17 KB